CN104376380B - A kind of ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network - Google Patents

A kind of ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network Download PDF

Info

Publication number
CN104376380B
CN104376380B CN201410655729.2A CN201410655729A CN104376380B CN 104376380 B CN104376380 B CN 104376380B CN 201410655729 A CN201410655729 A CN 201410655729A CN 104376380 B CN104376380 B CN 104376380B
Authority
CN
China
Prior art keywords
neuron
hidden layer
ammonia nitrogen
output
nitrogen concentration
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201410655729.2A
Other languages
Chinese (zh)
Other versions
CN104376380A (en
Inventor
韩红桂
李颖
张弛
张一弛
乔俊飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing University of Technology
Original Assignee
Beijing University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing University of Technology filed Critical Beijing University of Technology
Priority to CN201410655729.2A priority Critical patent/CN104376380B/en
Publication of CN104376380A publication Critical patent/CN104376380A/en
Priority to US14/668,836 priority patent/US9633307B2/en
Application granted granted Critical
Publication of CN104376380B publication Critical patent/CN104376380B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/082—Learning methods modifying the architecture, e.g. adding, deleting or silencing nodes or connections
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F3/00—Biological treatment of water, waste water, or sewage
    • C02F3/006—Regulation methods for biological treatment
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/18—Water
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Energy or water supply
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2101/00—Nature of the contaminant
    • C02F2101/10—Inorganic compounds
    • C02F2101/16—Nitrogen compounds, e.g. ammonia
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/001—Upstream control, i.e. monitoring for predictive control
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/005—Processes using a programmable logic controller [PLC]
    • C02F2209/006—Processes using a programmable logic controller [PLC] comprising a software program or a logic diagram
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/04—Oxidation reduction potential [ORP]
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/10—Solids, e.g. total solids [TS], total suspended solids [TSS] or volatile solids [VS]
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/14—NH3-N
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/18—PO4-P
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/22—O2
    • C—CHEMISTRY; METALLURGY
    • C02—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F—TREATMENT OF WATER, WASTE WATER, SEWAGE, OR SLUDGE
    • C02F2209/00—Controlling or monitoring parameters in water treatment
    • C02F2209/40—Liquid flow rate
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
    • G01N33/18—Water
    • G01N33/188—Determining the state of nitrification

Landscapes

  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Chemical & Material Sciences (AREA)
  • Water Supply & Treatment (AREA)
  • Business, Economics & Management (AREA)
  • Organic Chemistry (AREA)
  • Environmental & Geological Engineering (AREA)
  • Hydrology & Water Resources (AREA)
  • Microbiology (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Economics (AREA)
  • Pathology (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Analytical Chemistry (AREA)
  • Medicinal Chemistry (AREA)
  • Food Science & Technology (AREA)
  • Human Resources & Organizations (AREA)
  • Public Health (AREA)
  • Marketing (AREA)
  • Primary Health Care (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)

Abstract

A kind of ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network both belongs to control field, and water treatment field is belonged to again.For current sewage disposal process water outlet ammonia nitrogen concentration measurement process is cumbersome, instrument and equipment cost high, measurement result reliability and the low problem of accuracy, the present invention is based on municipal sewage treatment biochemical reaction characteristic, the prediction to crucial water quality parameter ammonia nitrogen concentration is realized using a kind of recurrence self organizing neural network, the problem of water outlet ammonia nitrogen concentration is difficult to measurement is solved；As a result show that the recurrence self organizing neural network can quickly and accurately predict the concentration of sewage disposal water outlet ammonia nitrogen, be conducive to lifting sewage processing procedure water outlet ammonia nitrogen concentration quality monitoring level and strengthen municipal sewage plant's fine-grained management.

Description

A kind of ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network
Technical field
The present invention is based on sewage disposal biochemical reaction characteristic, utilizes a kind of recurrence self organizing neural network of sensitivity analysis The prediction to the crucial water quality parameter ammonia nitrogen concentration of sewage disposal process is realized, ammonia nitrogen concentration is to characterize water pollution and sewage disposal The Important Parameters of degree, to health important, realize that the on-line prediction of ammonia nitrogen concentration realizes that denitrogenation is controlled Basic link, is the important branch in advanced manufacturing technology field, both belongs to control field, water treatment field is belonged to again.
Background technology
Ammonia nitrogen is the principal element of water environment pollution and body eutrophication problem, control water environment pollution and water body richness battalion One important measure of fosterization is exactly the discharge of ammonia nitrogen in strict limitation sewage disposal water outlet；During " 12 ", ammonia nitrogen concentration The restrictive Con trolling index of national discharge of major pollutant is turned into, ammonia nitrogen concentration intelligent testing technology can improve ammonia nitrogen removal Efficiency, improves the exceeded phenomenon of current water outlet ammonia nitrogen；Be conducive to lifting real-time water quality quality monitoring level and strengthen municipal sewage Treatment plant's fine-grained management, not only with preferable economic benefit, and with significant environmental and social benefits.Therefore, originally The achievement in research of invention has broad application prospects.
Environmental Protection Department issue《2013 China Environmental State Bulletins》In point out, China Huanghe valley, preserved egg in 2013 The large watershed of river basin etc. four and provincial boundaries water body are by different degrees of ammonia and nitrogen pollution, reservoir eutrophy, the middle nutrient laden ratio such as lake Example up to 95.2%.And ammonia nitrogen is to cause the key factor of body eutrophication, ammonia and nitrogen pollution on quality of water environment into For nationwide pollution problem；Therefore, the fast prediction of ammonia nitrogen concentration is realized, sewage disposal plant effluent ammonia nitrogen row up to standard is controlled Put, be the necessary links for ensureing that sewage disposal plant effluent water quality is qualified；The measuring method of current ammonia nitrogen concentration mainly has light splitting light Degree method, electrochemical methods and mechanism model etc., and the measuring principle of AAS is by free state ammonia or ammonium ion in water Reacted with the alkaline solution of mercury chloride and KI and generate light red brown colloidal state complex compound, by the extinction for measuring complex compound Degree can draw the content of ammonia nitrogen；However, this method measurement error is larger, disturbing factor is more, cumbersome, there is discarded object peace The problems such as full processing；Electrode method need not be pre-processed to water sample, and colourity and turbidity influence smaller to measurement result, be difficult by It is swift to operate simple to interference, but life-span and the less stable of electrode, meanwhile, electrode method measurement accuracy is relatively low；Meanwhile, it is dirty Water treatment procedure influence nitration reaction parameter is numerous, and dynamics is complicated, and then influences the parameter of ammonia nitrogen concentration numerous, it is each because Plain interphase interaction, the features such as non-linear and coupling is presented, it is difficult to set up the mechanism model of water outlet ammonia nitrogen；Therefore, it is existing Ammonia nitrogen concentration detection method is difficult to meet the demand that sewage treatment plant is detected in real time, it is necessary to seek new detection method；In recent years, With the development of soft-measuring technique, flexible measurement method can realize the Prediction of Nonlinear Dynamical Systems in certain accuracy rating, be ammonia nitrogen Concentration prediction provides theoretical foundation, and a kind of feasible method is provided for the high-precision forecast of ammonia nitrogen concentration.
The present invention devises a kind of water outlet ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network, realizes water outlet The on-line prediction of ammonia nitrogen concentration.
The content of the invention
Present invention obtains a kind of water outlet ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network, pass through design Recurrence self organizing neural network, according to the data of the real-time collection of sewage disposal process realize recurrence self organizing neural network Line is corrected, and realizes the real-time measurement of water outlet ammonia nitrogen concentration, is solved sewage disposal process water outlet ammonia nitrogen concentration and is difficult to real-time survey The problem of amount, the real-time monitoring level of municipal sewage plant's water quality is improved, ensure that sewage disposal process is normally run；
Present invention employs following technical scheme and realize step：
A kind of water outlet ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network comprises the following steps：
(1) auxiliary variable is determined：The actual water quality parameter data of sewage treatment plant are gathered, are chosen related to water outlet ammonia nitrogen concentration The strong water quality variable of property：Water inlet total phosphorus TP, the terminal oxidized reduction potential ORP of anaerobism, aerobic leading portion dissolved oxygen DO, aerobic end are total The auxiliary variable that solid suspension TSS and water outlet pH are predicted as water outlet ammonia nitrogen concentration；
(2) the recurrence self organizing neural network topological structure predicted designed for water outlet ammonia nitrogen concentration, recurrence self-organizing god It is divided into three layers through network：Input layer, hidden layer, output layer；Initialize recurrence self organizing neural network：Determine neutral net 5-K- 1 connected mode, i.e. input layer are 5, and hidden layer neuron is K, and K is positive integer, and output layer neuron is 1 It is individual；Parameter to neutral net carries out assignment；If having T training sample, the input of t neutral net is u (t)=[u1 (t), u2(t), u3(t), u4(t), u5(t)], the desired output of neutral net is expressed as yd(t), reality output is expressed as y (t)； The computing function of recurrence self organizing neural network is：
Represent the connection weight of k-th of neuron of t hidden layer and output layer, k=1,2 ..., K；vk(t) It is the output of k-th of neuron of t hidden layer, its calculation formula is：
Represent the connection weight of k-th of neuron of m-th of neuron of t input layer and hidden layer, m=1, 2 ..., 5；The self feed back output of k-th of hidden layer neuron of t is represented, its calculation formula is：
Represent the self-feedback connection weights value of k-th of neuron of t hidden layer, vk(t-1) it is the t-1 moment The output of k-th of neuron of hidden layer；
Defining error function is：
T represents the number of training of recurrence self organizing neural network input；
(3) neutral net is trained, is specially：
1. the recurrent neural network that a hidden layer neuron is K is given, input training sample data u (t) is initialized hidden Connection weight containing layer and output layerInitialize the self-feedback connection weights value of hidden layer neuronInitialization is defeated Enter the connection weight of layer and hidden layerM=1,2 ..., 5, k=1,2 ..., K；WithJust Initial value can take the Arbitrary Digit of (0,1)；Expected error value is set to Ed, Ed∈(0,0.01]；
2. the sensitivity of k-th of hidden layer neuron is calculated：
Wherein, k=1,2 ..., K；
Vark[E(y(t)|vk(t))]=2 (Ak)2+(Bk)2；
AkAnd BkThe fourier coefficient of sensitivity analysis is represented, its calculation formula is：
Wherein, Fourier variable s span is [- π, π]；ωk(t) be k-th of hidden layer neuron specified frequency Rate, ωk(t) determined by the output of k-th of hidden layer neuron：
bk(t) be trained t step in k-th of hidden layer neuron output maximum, ak(t) it is during the t trained is walked The output minimum value of k-th of hidden layer neuron；
3. neural network structure adjustment is carried out：
Delete adjustment：If the sensitivity S T of k-th of hidden layer neuronkLess than α1, α1∈ (0,0.01], then delete the god Through member, and hidden layer neuron number is updated for K1=K-1；Otherwise, the neuron, K are not deleted1=K；
Increase adjustment：If current error E (t)>Ed, then at the beginning of increasing a hidden layer neuron, the neuron newly inserted Beginning connection weight is：
Wherein,The connection weight between new insertion neuron and input layer is represented,Represent new insertion god Self-feedback connection weights value through member,The connection weight of new insertion neuron and output layer is represented, neuron h is hidden layer In the maximum neuron of sensitivity,Represent the connection weight of h-th of neuron of hidden layer and input layer before structural adjustment Value,The connection weight of h-th of neuron of hidden layer and output layer before structural adjustment is represented, and newly inserts neuron Export vnew(t) it is expressed as：
It is K to update hidden layer neuron number2=K1+1；Otherwise, the structure of neutral net, K are not adjusted2=K1；
4. neutral net connection weight adjustment is carried out：
Wherein, k=1,2 ..., K2；η1∈(0,0.1]、η2∈(0,0.1] And η3∈ (0,0.01] input layer and learning rate, the hidden layer neuron self-feedback connection weights of hidden layer connection weight are represented respectively The learning rate and hidden layer of value and the learning rate of output layer connection weight；
5. input training sample data x (t+1), repeat step 2. -4., the training of all training samples stops meter after terminating Calculate；
(4) test sample data are regard as the input of the recurrence self organizing neural network after training, recurrence self-organizing nerve The output of network is the predicted value of water outlet ammonia nitrogen concentration.
The creativeness of the present invention is mainly reflected in：
(1) for current sewage disposal plant effluent ammonia nitrogen concentration can not measure in real time the problem of, the present invention by extract with 5 related correlated variables of water outlet ammonia nitrogen concentration：Water inlet total phosphorus TP, the terminal oxidized reduction potential ORP of anaerobism, the dissolving of aerobic leading portion Oxygen DO, aerobic end total solid suspension TSS and water outlet pH, it is proposed that a kind of water outlet based on recurrence self organizing neural network Ammonia nitrogen concentration Forecasting Methodology, realizes the prediction of water outlet ammonia nitrogen concentration, and solve that water outlet ammonia nitrogen concentration is difficult to measure in real time asks Topic.
(2) present invention is the process of a complicated, dynamic time-varying, water outlet ammonia nitrogen concentration according to current sewage disposal process The features such as relation between correlated variables not only has non-linear, close coupling, and be difficult to be described with mathematical models, because This, based on actual sewage treatment plant measured data, employing recurrence self organizing neural network realizes the pre- of water outlet ammonia nitrogen concentration Survey, it is high with precision of prediction, there is the features such as well adapting to ability to environmental difference；
It is important to note that：The present invention uses 5 correlated variables related to water outlet ammonia nitrogen concentration, based on recurrence self-organizing god A kind of Forecasting Methodology of water outlet ammonia nitrogen concentration through network design, as long as the correlated variables and method that employ the present invention are gone out The prediction of water ammonia nitrogen concentration should all belong to the scope of the present invention.
Brief description of the drawings
Fig. 1 is the water outlet ammonia nitrogen concentration Forecasting Methodology structure chart of the present invention；
Fig. 2 is the water outlet ammonia nitrogen concentration Forecasting Methodology training result figure of the present invention；
Fig. 3 is the water outlet ammonia nitrogen concentration Forecasting Methodology training error figure of the present invention；
The water outlet ammonia nitrogen concentration that Fig. 4 is the present invention predicts the outcome figure；
Fig. 5 is the water outlet ammonia nitrogen concentration prediction-error image of the present invention.
Embodiment
Present invention obtains a kind of water outlet ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network, pass through design Recurrence self organizing neural network, according to the data of the real-time collection of sewage disposal process realize recurrence self organizing neural network Line is corrected, and realizes the real-time measurement of water outlet ammonia nitrogen concentration, is solved sewage disposal process water outlet ammonia nitrogen concentration and is difficult to real-time survey The problem of amount, the real-time monitoring level of municipal sewage plant's water quality is improved, ensure that sewage disposal process is normally run；
Experimental data is from certain sewage treatment plant annual water analysis daily sheet in 2014；Take into water total phosphorus TP, detest respectively The terminal oxidized reduction potential ORP of oxygen, aerobic leading portion dissolved oxygen DO, aerobic end total solid suspension TSS, water outlet pH and water outlet ammonia The actually detected data of nitrogen concentration are remaining 245 groups of data availables after experiment sample data, rejecting abnormalities experiment sample, will all 245 groups of data samples be divided into two parts：Wherein 165 groups data are used as test sample as training sample, remaining 80 groups of data；
1. a kind of water outlet ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network comprises the following steps：
(1) auxiliary variable is determined：The actual water quality parameter data of sewage treatment plant are gathered, are chosen related to water outlet ammonia nitrogen concentration The strong water quality variable of property：Water inlet total phosphorus TP, the terminal oxidized reduction potential ORP of anaerobism, aerobic leading portion dissolved oxygen DO, aerobic end are total The auxiliary variable that solid suspension TSS and water outlet pH are predicted as water outlet ammonia nitrogen concentration；
(2) the recurrence self organizing neural network topological structure predicted designed for water outlet ammonia nitrogen concentration, recurrence self-organizing god It is divided into three layers through network：Input layer, hidden layer, output layer；Initialize recurrence self organizing neural network：Determine neutral net 5-3- 1 connected mode, i.e. input layer are 5, and hidden layer neuron is 3, and output layer neuron is 1, such as Fig. 1；To god Parameter through network carries out assignment；If the input of t neutral net is u (t)=[u1(t), u2(t), u3(t), u4(t), u5 (t)], the desired output of neutral net is expressed as yd(t), reality output is expressed as y (t), has T training sample；Recurrence is certainly Organizing the computing function of neutral net is：
Represent the connection weight of k-th of neuron of t hidden layer and output layer, k=1,2,3；vk(t) it is The output of k-th of neuron of t hidden layer, its calculation formula is：
Represent the connection weight of k-th of neuron of i-th of neuron of t input layer and hidden layer, m=1, 2 ..., 5；The self feed back output of k-th of hidden layer neuron of t is represented, its calculation formula is：
Represent the self-feedback connection weights value of k-th of neuron of t hidden layer, vk(t-1) it is the t-1 moment The output of k-th of neuron of hidden layer；
Defining error function is：
T represents the number of training of recurrence self organizing neural network input；
(3) neutral net is trained, is specially：
1. the recurrent neural network that a hidden layer neuron is K is given, input training sample data u (t) is initialized hidden Connection weight containing layer and output layerInitialize the self-feedback connection weights value of hidden layer neuronInitialization is defeated Enter the connection weight of layer and hidden layerM=1,2 ..., 5, k=1,2 ..., K；WithJust Initial value can take the Arbitrary Digit of (0,1)；Expected error value is set to Ed=0.01；
2. the sensitivity of k-th of hidden layer neuron is calculated：
Wherein, k=1,2 ..., K；
Vark[E(y(t)|vk(t))]=2 (Ak)2+(Bk)2；
AkAnd BkThe fourier coefficient of sensitivity analysis is represented, its calculation formula is：
Wherein, Fourier variable s span is [- π, π]；ωk(t) be k-th of hidden layer neuron specified frequency Rate, ωk(t) determined by the output of k-th of hidden layer neuron：
bk(t) be trained t step in k-th of hidden layer neuron output maximum, ak(t) it is during the t trained is walked The output minimum value of k-th of hidden layer neuron；
3. neural network structure adjustment is carried out：
Delete adjustment：If the sensitivity S T of k-th of hidden layer neuronkLess than α1=0.01, then the neuron is deleted, and It is K to update hidden layer neuron number1=K-1；Otherwise, the neuron, K are not deleted1=K；
Increase adjustment：If current error E (t)>Ed, then at the beginning of increasing a hidden layer neuron, the neuron newly inserted Beginning connection weight is：
Wherein,The connection weight between new insertion neuron and input layer is represented,Represent new insertion god Self-feedback connection weights value through member,The connection weight of new insertion neuron and output layer is represented, neuron h is hidden layer In the maximum neuron of sensitivity,Represent the connection weight of h-th of neuron of hidden layer and input layer before structural adjustment Value,The connection weight of h-th of neuron of hidden layer and output layer before structural adjustment is represented, and newly inserts neuron Export vnew(t) it is expressed as：
It is K to update hidden layer neuron number2=K1+1；Otherwise, the structure of neutral net, K are not adjusted2=K1；
4. neutral net connection weight adjustment is carried out：
Wherein, k=1,2 ..., K2；Input layer and hidden layer connection weight The learning rate of value, the learning rate of hidden layer neuron self-feedback connection weights value and hidden layer and output layer connection weight Habit rate is respectively η1=0.01, η2=0.01, η3=0.005；
5. input training sample data x (t+1), repeat step 2. -4., the training of all training samples stops meter after terminating Calculate；
The training result of recurrence self organizing neural network is as shown in Fig. 2 X-axis：Sample number, unit is individual/sample, Y-axis：Go out Water ammonia nitrogen concentration, unit is mg/L, and solid line is water outlet ammonia nitrogen concentration real output value, and dotted line is that recurrence self organizing neural network is defeated Go out value；Error such as Fig. 3 of water outlet ammonia nitrogen concentration real output value and recurrence self organizing neural network output valve, X-axis：Sample number, Unit is individual/sample, Y-axis：Water outlet ammonia nitrogen concentration, unit is mg/L；
(4) test sample data are regard as the input of the recurrence self organizing neural network after training, recurrence self-organizing nerve The output of network is water outlet ammonia nitrogen concentration value；Predict the outcome as shown in figure 4, X-axis：Sample number, unit is individual/sample, Y-axis： Water outlet ammonia nitrogen concentration, unit is mg/L, and solid line is water outlet ammonia nitrogen concentration real output value, and dotted line is that the prediction of water outlet ammonia nitrogen concentration is defeated Go out value；Water outlet ammonia nitrogen concentration real output value predicts error such as Fig. 5 of output valve, X-axis with water outlet ammonia nitrogen concentration：Sample number, it is single Position is individual/sample, Y-axis：Water outlet ammonia nitrogen concentration predicated error, unit is mg/L；As a result show to be based on recurrence self-organizing feature map The validity of the water outlet ammonia nitrogen concentration Forecasting Methodology of network.
Table 1-14 is experimental data of the present invention, and table 1-6 is training sample：The terminal oxidized reduction electricity of total phosphorus TP, the anaerobism of intaking Position ORP, aerobic leading portion dissolved oxygen DO, aerobic end total solid suspension TSS, water outlet pH and actual measurement water outlet ammonia nitrogen concentration, table 7 is The output of recurrence self organizing neural network in training process, table 8-14 is test sample：Total phosphorus TP, the anaerobism of intaking are terminal oxidized also Former current potential ORP, aerobic leading portion dissolved oxygen DO, aerobic end total solid suspension TSS, water outlet pH and actual measurement water outlet ammonia nitrogen concentration, Table 14 is water outlet ammonia nitrogen concentration predicted value of the present invention.
Training sample
The auxiliary variable of table 1. water inlet total phosphorus TP (mg/L)
3.9021 3.8943 4.3182 4.2219 4.6025 4.3496 4.5057 4.5057 4.5057 4.5057
3.8848 3.8155 3.9287 4.0154 4.1802 4.1465 4.1465 4.1465 4.1465 4.1465
4.1465 4.1465 4.2845 3.8326 3.7941 4.4504 4.3140 4.4706 4.2410 4.5929
4.4944 3.8420 3.8664 4.0551 4.2081 4.1305 4.2712 3.5370 2.8337 4.1774
3.7040 3.6206 4.1277 4.0534 4.3345 4.1899 4.3530 4.2267 4.1365 4.0805
4.0221 3.9322 3.8749 4.0820 4.0727 4.1665 4.2180 4.1436 4.3808 4.4049
4.2351 4.2345 4.1325 3.9768 3.9608 3.7857 3.8670 3.8294 3.9176 4.0762
4.0099 4.1032 4.0226 4.0941 4.1105 4.1284 4.0332 4.0053 3.9005 3.8975
3.7953 3.8648 3.8835 3.9725 4.2412 4.4562 4.2018 4.1647 4.5131 4.1541
4.0418 4.0789 3.9439 3.7140 3.9232 4.0274 3.9716 4.0438 4.2394 4.2394
4.2394 4.2394 4.2394 4.2394 4.2394 4.2394 4.2394 4.2392 4.2392 4.2392
4.2392 4.2392 4.2392 4.2392 3.6244 4.2873 4.0612 3.9821 4.0342 4.0920
4.0371 4.0575 4.1273 4.1907 4.2153 4.2907 4.1859 4.1446 4.0744 4.3648
3.8792 3.7862 3.8169 3.7380 3.8215 4.0155 4.0076 3.9549 4.0678 4.0160
3.9320 4.0386 3.9331 3.8880 3.7802 3.6751 3.6112 3.6098 3.6671 3.6269
3.7581 3.8980 4.0578 3.9783 3.9331 3.9794 4.0795 4.1422 4.7669 4.3334
4.4615 4.1052 4.0354 4.0672 4.2935
The terminal oxidized reduction potential ORP of the auxiliary variable anaerobism of table 2.
-540.2970 -546.8350 -554.3970 -556.1280 -553.4360 -551.0650 -549.9110 -554.5260 -556.3200 -561.1910
-555.0380 -548.5010 -550.9370 -563.9470 -564.7160 -565.5500 -565.2290 -565.1010 -563.7550 -564.7160
-564.7800 -565.6140 -565.6140 -564.6520 -563.8830 -566.1260 -565.5500 -565.2290 -564.8440 -470.6930
-480.6910 -414.3560 -539.2080 -555.3590 -557.9870 -558.8200 -558.9480 -526.9660 -470.4370 -567.4720
-565.1650 -563.8190 -578.6880 -581.2520 -581.2520 -582.1490 -581.8290 -581.7650 -581.7650 -581.2520
-580.9960 -580.4830 -579.8420 -579.7140 -579.7780 -580.4190 -580.7390 -580.5470 -580.7390 -580.4830
-580.0980 -580.0340 -579.0730 -578.6240 -578.2400 -578.3040 -576.9580 -577.4070 -577.9830 -578.2400
-578.1760 -577.8550 -577.7270 -577.4710 -577.2780 -577.0860 -576.8940 -576.8940 -577.4070 -575.0350
-572.9840 -573.7530 -574.9070 -574.7790 -575.0990 -575.2270 -573.8170 -572.2150 -572.6640 -573.0480
-572.4070 -572.0230 -571.4460 -573.9460 -573.8170 -573.9460 -574.2660 -574.9070 -575.7400 -575.7400
-575.0990 -575.0350 -574.5860 -574.1380 -573.9460 -573.6890 -573.3050 -575.0350 -574.8430 -574.1380
-574.0100 -573.6890 -573.7530 -572.0230 -570.9970 -570.1000 -569.9720 -570.7410 -571.7020 -572.1510
-572.1510 -572.6640 -573.2410 -573.3690 -573.1760 -573.1120 -573.0480 -573.0480 -573.1120 -573.1760
-574.5860 -578.3680 -578.7530 -577.2780 -573.2410 -570.4210 -570.9330 -572.0870 -572.1510 -570.2920
-570.0360 -568.7540 -567.0240 -568.4340 -569.0100 -568.8820 -568.9460 -569.2670 -569.4590 -569.5230
-570.1000 -571.5100 -572.4070 -572.8560 -572.1510 -570.6770 -570.3560 -569.9720 -569.6510 -569.5870
-569.7150 -570.1640 -570.8690 -570.9330 -571.5740
The aerobic leading portion dissolved oxygen DO (mg/L) of the auxiliary variable of table 3.
0.0518 0.0394 0.0379 0.0356 0.0370 0.0361 0.0467 0.0417 0.0510 0.0382
0.0411 0.0363 0.0472 0.0581 0.0514 0.0561 0.0673 0.0585 0.0507 0.0486
0.0484 0.0492 0.1343 0.0793 0.0561 0.0696 0.0427 0.0441 0.0480 0.0571
0.0464 0.0425 0.0540 0.0711 0.0715 0.0535 0.0792 0.0603 0.0522 0.0375
0.0391 0.0382 0.0318 0.0339 0.0312 0.0831 0.0403 0.0353 0.0411 0.0355
0.0501 0.0384 0.0371 0.0962 0.0497 0.0666 0.0398 0.0427 0.0663 0.0416
0.0640 0.0555 0.0796 0.0768 0.0615 0.0592 0.0946 0.0530 0.0769 0.0450
0.0823 0.0397 0.0567 0.0390 0.0396 0.0716 0.0423 0.0637 0.0448 0.3747
0.3764 0.4340 0.4833 0.4329 0.4512 0.4455 0.5192 0.4821 0.4478 0.4694
0.4844 0.5815 0.5309 0.9670 0.8274 0.7756 0.4701 0.4711 0.4316 0.4357
0.4621 0.4867 0.5287 0.5043 0.5440 0.5487 0.5110 0.4867 0.4889 0.5043
0.5378 0.5487 0.5400 1.5057 1.0497 0.9117 0.9334 0.8063 0.4684 0.4649
0.4508 0.3812 0.3495 0.3594 0.3574 0.3821 0.3640 0.3554 0.3703 1.0503
0.7617 0.5861 0.5539 0.4448 0.2693 0.2558 0.2740 0.3096 0.2734 0.2962
0.2997 0.3444 0.3165 0.2646 0.2404 0.3987 0.3624 0.3024 0.3268 0.2476
0.2465 0.2079 0.2103 0.2380 0.2519 0.2651 0.2470 0.2557 0.2890 0.2659
0.9111 0.7375 0.2701 0.2665 0.2489
The aerobic end total solid suspension TSS (mg/L) of the auxiliary variable of table 4.
2.8251 2.7176 2.7700 2.8094 2.7666 2.7748 2.7823 2.7998 2.8015 2.7686
2.7556 2.7975 2.8011 2.8182 2.8985 2.8089 2.7813 2.8060 3.1727 2.9242
2.8536 2.8202 2.8179 2.9067 2.7963 2.8271 2.8168 2.8262 2.8678 2.8074
2.8428 2.8260 2.8615 2.7277 2.7863 2.8132 2.7385 2.8738 2.8651 2.9005
2.9324 2.8942 2.8223 2.8512 2.7712 2.6251 2.5540 2.4976 2.6220 2.6049
2.5314 2.5817 2.5765 2.5590 2.5611 2.5664 2.5177 2.4709 2.4971 2.4192
2.4831 2.5234 2.4654 2.4501 2.4564 2.4367 2.4777 2.4562 2.4776 2.4068
2.4583 2.4031 2.4443 2.5130 2.4505 2.4376 2.3933 2.4439 2.4637 2.4573
2.4982 2.5214 2.4515 2.3733 2.4492 2.4602 2.4725 2.4949 2.4815 2.5655
2.5286 2.4330 2.4429 2.4573 2.4820 2.6305 2.5025 2.4821 2.4912 2.4121
2.4265 2.4700 2.4481 2.4801 2.5045 2.4743 2.4331 2.4700 2.3919 2.4801
2.4472 2.4743 2.4740 2.5777 2.4818 2.5754 2.5450 2.5624 2.5353 2.4304
2.3899 2.3654 2.4347 2.3155 2.3089 2.2740 2.3947 2.2430 2.3166 2.2692
2.2754 2.3157 2.2768 2.1761 2.2200 2.1312 2.3333 2.4261 2.4155 2.3439
2.3083 2.3119 2.2717 2.2823 2.4388 2.4274 2.5251 2.4161 2.4789 2.3514
2.3938 2.2736 2.3829 2.3818 2.4428 2.4255 2.3938 2.4187 2.5133 2.4147
2.5321 2.4440 2.3300 2.2835 2.4055
The auxiliary variable water outlet pH of table 5.
7.9266 7.9298 7.9266 7.9176 7.8907 7.8718 7.8641 7.8520 7.8465 7.8448
7.8536 7.8579 7.8643 7.8643 7.8655 7.8645 7.8623 7.8568 7.8581 7.8595
7.8619 7.8632 7.8690 7.8713 7.8801 7.9154 7.9079 7.9038 7.9029 7.9466
7.9524 7.8931 7.9049 7.9176 7.9166 7.9110 7.8953 7.8901 7.8949 8.0150
8.0054 8.0039 7.9967 8.0228 7.9988 7.9917 7.9863 7.9852 7.9898 7.9908
7.9962 7.9949 7.9981 8.0005 7.9996 8.0042 8.0112 8.0102 8.0000 7.9967
7.9946 7.9947 7.9856 7.9844 7.9933 7.9970 7.9909 8.0009 8.0056 8.0036
8.0003 7.9993 8.0028 8.0065 8.0043 8.0035 8.0025 8.0028 8.0041 8.0044
8.0137 8.0184 8.0276 8.0242 8.0302 8.0337 8.0225 7.9939 8.0150 8.0210
8.0272 8.0274 8.0278 8.0275 8.0334 8.0398 8.0430 8.0443 8.0403 8.0348
8.0261 8.0217 8.0151 8.0088 8.0128 8.0119 7.9982 8.0217 8.0184 8.0088
8.0091 8.0119 8.0132 7.9865 7.9966 8.0214 8.0305 8.0523 8.0649 8.0616
8.0617 8.0597 8.0542 8.0328 8.0260 8.0137 8.0140 8.0108 8.0097 8.0142
8.0106 8.0296 8.0339 8.0221 8.0095 8.0303 8.0385 8.0399 8.0412 8.0335
8.0279 8.0111 7.9768 8.0001 8.0139 8.0204 8.0164 8.0153 8.0182 8.0221
8.0277 8.0347 8.0314 8.0202 8.0157 8.0092 8.0107 8.0097 8.0146 8.0159
8.0146 8.0166 8.0448 8.0585 8.0826
The actual measurement water outlet ammonia nitrogen concentration of table 6. (mg/L)
3.7214 3.6922 3.3211 3.3147 3.3754 3.4273 3.4585 3.5697 3.5634 3.6763
3.7086 3.6714 3.8618 3.6722 3.5585 3.6395 3.5802 3.6442 3.7178 3.8003
3.8684 3.9189 3.8830 3.8383 3.8612 3.6437 3.6019 3.6432 3.7056 3.6175
3.5967 3.5521 3.5992 3.5789 3.6120 3.5846 3.5920 3.5888 3.5520 3.7352
3.8218 3.9312 5.8870 5.7259 7.5603 11.9231 12.1773 12.2836 12.3372 12.3155
12.4116 12.5365 12.4893 12.2718 12.4335 12.3200 12.3238 12.3038 12.5816 12.4523
12.5137 12.7659 12.9055 12.7696 12.8395 13.1354 12.8835 12.9153 13.0054 12.9308
12.9644 13.0146 12.9466 13.1046 13.0941 13.0794 13.2232 13.1832 13.1733 13.2032
12.8992 12.7643 12.4099 12.2235 11.7775 11.5723 11.3341 11.2749 11.0900 10.9602
10.7810 10.7283 10.6037 9.6868 9.1768 8.9925 8.5913 8.5682 8.4254 8.3490
8.2571 8.2967 8.2521 8.1850 8.1911 8.1174 8.0427 8.2967 8.3094 8.1850
8.1843 8.1174 8.2504 7.9622 7.7317 7.4507 7.3742 6.9528 6.7038 6.3957
6.3379 6.3166 6.3299 6.5581 6.6947 7.0927 7.2973 7.7820 8.1116 9.0352
8.7383 8.7475 8.7663 8.7660 8.8353 8.8457 9.0967 9.3701 9.3140 9.0599
9.1053 9.2407 9.2865 9.3157 9.2816 9.3850 9.2125 8.9531 8.8280 8.5461
8.3717 8.1966 7.6552 9.3499 9.2675 9.2230 9.2480 9.3684 9.3754 9.2173
9.1306 8.8445 7.5305 7.1104 6.5671
The recurrence self organizing neural network of table 7. training output (mg/L)
3.7842 3.6955 3.4035 3.1000 3.4514 3.5299 3.4003 3.3512 3.6258 3.7517
3.6886 3.5692 4.0296 3.7519 3.7126 3.8381 3.8528 3.2400 3.6796 3.8111
3.8598 3.7948 3.7933 3.8403 3.8687 3.8490 3.4309 3.5505 3.5864 3.6058
3.6033 3.5463 3.4731 3.4313 3.5456 4.0032 3.7263 3.6194 3.5477 3.7518
3.8272 3.9173 5.8444 5.7479 7.5665 11.8148 12.1198 12.3419 12.3674 12.2235
12.4819 12.5644 12.5330 12.3365 12.4200 12.3603 12.4098 12.2661 12.5402 12.4473
12.5159 12.8958 12.7052 12.9661 13.0068 13.1035 12.6238 13.1129 12.6902 12.9062
12.7613 13.1369 13.0705 13.0488 13.2949 12.9133 12.9525 13.0572 13.3742 13.3882
12.7594 12.8822 12.4131 12.0293 10.3936 11.5563 10.6390 11.0043 11.1370 11.1234
10.2559 11.0945 10.4768 9.7053 9.1992 9.0008 8.7348 8.8083 8.5365 8.6181
8.7525 8.7364 8.0552 8.3347 8.3500 8.1183 8.1562 8.7374 8.2457 8.3358
7.8597 8.1193 8.3285 7.9669 7.7341 7.4802 7.3175 6.9490 6.6141 6.4781
6.4584 6.8932 6.6881 6.6964 6.8403 6.9678 7.4339 7.9103 7.9315 9.0342
8.7332 8.6464 8.8931 8.7614 8.8156 8.7724 8.7067 9.3423 8.3514 8.7110
9.2627 9.1725 9.3982 9.2134 8.9834 9.4617 9.1733 8.8833 9.0942 8.7205
8.0875 8.3975 8.1150 9.2811 9.7516 9.6039 9.3939 9.1582 9.3664 9.3410
9.1240 8.9294 7.7141 7.1464 6.5466
Test sample
The auxiliary variable of table 8. water inlet total phosphorus TP (mg/L)
3.9522 4.1867 4.5942 4.5057 4.5057 4.0066 3.7529 4.1116 4.1465 4.1465
4.1465 4.0993 4.2017 4.5199 4.1266 4.2198 3.4877 4.7860 3.9951 4.3522
4.4541 4.1859 4.2168 3.9868 3.9029 4.0702 4.1378 4.3289 4.3061 4.0605
4.1268 3.9708 3.9485 4.0112 4.1164 4.3104 4.0388 3.8027 3.7678 4.0382
4.2339 4.2524 4.1057 3.9310 3.9415 3.8455 4.3598 4.2394 4.2394 4.2394
4.2394 4.2392 4.2392 4.2392 4.2889 3.9926 4.1127 4.0208 4.1534 4.2663
4.2058 4.0359 3.8457 3.7628 3.9413 4.0122 3.9671 3.9380 3.9573 3.7158
3.6388 3.6132 3.8164 3.9993 3.9670 4.0034 4.1387 4.1678 3.9797 4.2248
The terminal oxidized reduction potential ORP of the auxiliary variable anaerobism of table 9.
-552.1540 -556.8970 -551.9620 -552.6030 -558.6280 -561.4480 -543.7580 -565.7420 -565.0370 -564.2680
-565.2930 -564.3960 -565.9980 -489.0880 -558.1790 -558.8200 -487.6130 -568.9460 -565.5500 -580.4190
-581.5730 -581.7010 -582.0850 -581.1880 -580.0980 -579.7780 -580.6110 -580.7390 -580.1630 -579.8420
-578.1120 -579.3930 -578.2400 -578.1760 -577.5990 -577.0860 -576.7020 -573.6890 -574.7150 -574.7790
-575.0350 -572.1510 -572.8560 -571.6380 -573.7530 -574.0100 -575.3560 -575.2920 -574.8430 -574.0100
-573.7530 -574.5860 -573.9460 -573.3050 -570.4210 -570.0360 -572.0230 -572.2150 -573.4970 -572.9840
-572.9840 -573.3050 -577.3420 -578.5600 -570.6130 -571.5740 -570.9970 -569.9080 -567.6650 -569.0100
-569.3310 -569.4590 -570.6130 -572.8560 -571.4460 -570.6130 -569.8440 -569.3950 -570.2920 -571.1900
The aerobic leading portion dissolved oxygen DO (mg/L) of the auxiliary variable of table 10.
0.0383 0.0428 0.0361 0.0378 0.0395 0.0602 0.0706 0.0453 0.0743 0.0735
0.0567 0.1172 0.0582 0.0398 0.0609 0.0811 0.0686 0.0398 0.0474 0.0317
0.0298 0.1265 0.0659 0.0971 0.0345 0.0355 0.0457 0.0488 0.0412 0.0545
0.0765 0.0364 0.0406 0.0843 0.0464 0.0346 0.1481 0.4026 0.3942 0.4193
0.4073 0.4379 0.5426 0.5498 0.8550 0.4882 0.4207 0.4564 0.4889 0.5378
0.5400 0.5287 0.5440 0.5110 0.8817 0.8742 0.4291 0.4537 0.3765 0.3696
0.3782 0.3274 0.7197 0.5351 0.2611 0.3343 0.3412 0.3301 0.2746 0.2365
0.3272 0.2974 0.2066 0.1995 0.2546 0.2459 0.2654 0.2566 0.2232 0.2282
The aerobic end total solid suspension TSS (mg/L) of the auxiliary variable of table 11.
2.8343 2.8151 2.7787 2.7807 2.7539 2.7827 2.8063 2.8055 2.9044 2.8029
2.7963 2.8936 2.8786 2.8337 2.7973 2.7974 2.8266 2.8632 2.9151 2.7774
2.8432 2.7067 2.6005 2.6635 2.5869 2.5829 2.5363 2.5279 2.4897 2.4674
2.4916 2.5265 2.5397 2.4082 2.4903 2.3932 2.4240 2.4906 2.5340 2.3839
2.4320 2.3993 2.5394 2.5140 2.4693 2.4245 2.4605 2.4649 2.3919 2.4472
2.4740 2.4481 2.5045 2.4331 2.4866 2.5113 2.4309 2.3655 2.3883 2.2805
2.3078 2.2824 2.2668 2.2297 2.2105 2.4196 2.2935 2.3671 2.3100 2.3821
2.4491 2.5777 2.4440 2.4318 2.4089 2.4784 2.4254 2.4256 2.3243 2.3120
The auxiliary variable water outlet pH of table 12.
7.9298 7.9087 7.8818 7.8586 7.8445 7.8517 7.8622 7.8667 7.8590 7.8593
7.8643 7.8702 7.9216 7.9536 7.9188 7.9032 7.8936 8.0238 8.0090 7.9940
8.0011 8.0101 7.9908 7.9930 7.9959 7.9983 8.0112 8.0045 7.9968 7.9936
7.9866 8.0030 8.0069 7.9992 8.0040 8.0033 8.0015 8.0090 8.0264 8.0254
8.0373 8.0021 8.0281 8.0288 8.0305 8.0431 8.0480 8.0316 8.0184 8.0091
8.0132 8.0151 8.0128 7.9982 8.0055 8.0419 8.0627 8.0595 8.0498 8.0158
8.0107 8.0120 8.0195 8.0314 8.0187 8.0398 8.0368 8.0281 7.9850 8.0196
8.0101 8.0212 8.0334 8.0235 8.0123 8.0105 8.0145 8.0124 8.0209 8.0745
The actual measurement water outlet ammonia nitrogen concentration of table 13. (mg/L)
3.5761 3.3048 3.4170 3.5679 3.5392 3.9342 3.5926 3.5754 3.5805 3.7210
3.9394 3.9206 3.7720 3.5899 3.5946 3.5928 3.5704 3.6951 3.7283 6.8643
7.6531 9.9438 12.0870 12.4108 12.2645 12.2824 12.3406 12.3668 12.5197 12.6702
12.7935 13.0679 12.9323 12.9189 13.1193 13.2119 13.1942 13.0278 12.5932 12.0214
11.5033 11.1842 10.8915 10.6223 9.3917 8.7883 8.5280 8.2748 8.3094 8.1843
8.2504 8.2521 8.1911 8.0427 7.6784 7.1995 6.5172 6.3016 6.3704 6.7937
7.6118 8.3032 8.7825 8.7420 8.7893 9.5518 9.2179 9.1266 9.2621 9.2021
9.0655 8.6186 8.2710 7.5227 9.3176 9.1937 9.2926 9.0822 8.6282 6.8153
The recurrence self organizing neural network of table 14. prediction output (mg/L)
3.0054 2.9792 4.1867 4.9286 4.2662 5.2209 4.8830 5.9236 4.0377 5.6451
6.2735 6.2896 4.7227 2.5800 2.4380 4.1350 2.3930 11.5193 5.2214 6.0038
5.9712 13.0544 10.9030 19.2732 12.8016 12.1521 12.1938 11.9632 12.8526 13.2788
12.6482 13.3323 12.9681 12.9030 13.3655 16.1601 11.5984 13.6644 13.0311 12.4301
9.8375 9.5739 10.0693 7.9654 10.2654 12.5032 10.2643 9.0101 7.8697 6.7043
7.0017 6.9231 6.9281 7.3861 5.1751 5.5377 7.3165 8.5132 7.9163 6.5856
6.6081 7.9339 8.8676 6.0381 9.3639 9.1078 9.9013 9.6566 9.9644 8.8577
7.8352 6.3314 7.5965 9.2300 9.5224 10.2648 9.0901 9.1036 9.1942 4.3949

Claims (1)

1. a kind of water outlet ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network, it is characterised in that including following step Suddenly：
(1) auxiliary variable is determined：The actual water quality parameter data of sewage treatment plant are gathered, are chosen strong with water outlet ammonia nitrogen concentration correlation Water quality variable：Water inlet total phosphorus TP, the terminal oxidized reduction potential ORP of anaerobism, aerobic leading portion dissolved oxygen DO, aerobic end total solid The auxiliary variable that suspension TSS and water outlet pH are predicted as water outlet ammonia nitrogen concentration；
(2) the recurrence self organizing neural network topological structure predicted designed for water outlet ammonia nitrogen concentration, recurrence self-organizing feature map Network is divided into three layers：Input layer, hidden layer, output layer；Initialize self-organizing population-radial base neural net：Determine nerve net Network 5-K-1 connected mode, i.e. input layer are 5, and hidden layer neuron is K, and K is positive integer, output layer nerve Member is 1；Parameter to neutral net carries out assignment；If having T training sample, the input of t neutral net is u (t) =[u1(t), u2(t), u3(t), u4(t), u5(t)], the desired output of neutral net is expressed as yd(t), reality output is expressed as y (t)；The calculation formula of recurrence self organizing neural network is：
y ( t ) = Σ k = 1 K w k 3 ( t ) v k ( t ) ; - - - ( 1 )
Represent the connection weight of k-th of neuron of t hidden layer and output layer, k=1,2 ..., K；vk(t) it is The output of k-th of neuron of t hidden layer, its calculation formula is：
v k ( t ) = f ( Σ m = 1 5 w m k 1 ( t ) u m ( t ) + v k 1 ( t ) ) ; - - - ( 2 )
Represent the connection weight of k-th of neuron of m-th of neuron of t input layer and hidden layer, m=1,2 ..., 5；The self feed back output of k-th of hidden layer neuron of t is represented, its calculation formula is：
v k 1 ( t ) = w k 2 ( t ) v k ( t - 1 ) ; - - - ( 3 )
Represent the self-feedback connection weights value of k-th of neuron of t hidden layer, vk(t-1) it is to imply at the t-1 moment The output of k-th of neuron of layer；
Defining error function is：
E ( t ) = 1 2 T Σ t = 1 T ( y d ( t ) - y ( t ) ) 2 ; - - - ( 4 )
T represents the number of training of recurrence self organizing neural network input；
(3) neutral net is trained, is specially：
1. the recurrent neural network that a hidden layer neuron is K is given, input training sample data u (t) initializes hidden layer With the connection weight of output layerInitialize the self-feedback connection weights value of hidden layer neuronInitialize input layer and The connection weight of hidden layerM=1,2 ..., 5, k=1,2 ..., K；WithInitial value take (0,1) Arbitrary Digit；Expected error value is set to Ed, Ed∈(0,0.01]；
2. the sensitivity of k-th of hidden layer neuron is calculated：
ST k ( t ) = Var k [ E ( y ( t ) | v k ( t ) ) ] V a r [ y ( t ) ] ; - - - ( 5 )
Wherein, k=1,2 ..., K；
Vark[E(y(t)|vk(t))]=2 (Ak)2+(Bk)2；
V a r ( y ( t ) ) = 2 Σ k = 1 K ( ( A k ) 2 + ( B k ) 2 ) ; - - - ( 6 )
AkAnd BkThe fourier coefficient of sensitivity analysis is represented, its calculation formula is：
A k = 1 2 π ∫ - π π c o s ( ω k ( t ) s ) d s ;
B k = 1 2 π ∫ - π π s i n ( ω k ( t ) s ) d s ; - - - ( 7 )
Wherein, Fourier variable s span is [- π, π]；ωk(t) be k-th of hidden layer neuron assigned frequency, ωk (t) determined by the output of k-th of hidden layer neuron：
ω k ( t ) = a r c s i n π b k ( t ) - a k ( t ) ( v k ( t ) - b k ( t ) + a k ( t ) 2 ) ; - - - ( 8 )
bk(t) be trained t step in k-th of hidden layer neuron output maximum, ak(t) it is kth during the t trained is walked The output minimum value of individual hidden layer neuron；
3. neural network structure adjustment is carried out：
Delete adjustment：If the sensitivity S T of k-th of hidden layer neuronkLess than α1, α1∈ (0,0.01], then the neuron is deleted, And hidden layer neuron number is updated for K1=K-1；Otherwise, the neuron, K are not deleted1=K；
Increase adjustment：If current error E (t)>Ed, then a hidden layer neuron is increased, the neuron newly inserted is initially connected Weights are：
w n e w 1 ( t ) = w h 1 ( t ) = [ w 1 h 1 ( t ) , w 2 h 1 ( t ) , ... , w 5 h 1 ( t ) ] ;
w n e w 2 ( t ) = w h 2 ( t ) ;
w n e w 3 ( t ) = y d ( t ) - y ( t ) v n e w ( t ) ; - - - ( 9 )
Wherein,The connection weight between new insertion neuron and input layer is represented,Represent new insertion neuron Self-feedback connection weights value,The connection weight of new insertion neuron and output layer is represented, neuron h is the spirit in hidden layer The maximum neuron of sensitivity,The connection weight of h-th of neuron of hidden layer and input layer before structural adjustment is represented, The connection weight of h-th of neuron of hidden layer and output layer before structural adjustment is represented, and newly inserts the output v of neuronnew (t) it is expressed as：
v n e w ( t ) = f ( Σ m = 1 5 w m h 1 ( t ) u m ( t ) + v n e w 1 ( t ) ) ;
v n e w 1 ( t ) = w h 2 ( t ) v h ( t - 1 ) ; - - - ( 10 )
It is K to update hidden layer neuron number2=K1+1；Otherwise, the structure of neutral net, K are not adjusted2=K1；
4. neutral net connection weight adjustment is carried out：
w k 1 ( t + 1 ) = w k 1 ( t ) + η 1 ∂ E ( t ) ∂ w k 1 ( t ) ;
w k 2 ( t + 1 ) = w k 2 ( t ) + η 2 ∂ E ( t ) ∂ w k 2 ( t ) ;
w k 3 ( t + 1 ) = w k 3 ( t ) + η 3 ∂ E ( t ) ∂ w k 3 ( t ) ; - - - ( 11 )
Wherein, k=1,2 ..., K2；η1∈(0,0.1]、η2∈ (0,0.1] and η3 ∈ (0,0.01] learning rate of input layer and hidden layer connection weight, hidden layer neuron self-feedback connection weights value are represented respectively The learning rate of learning rate and hidden layer and output layer connection weight；
5. input training sample data u (t+1), repeat step 2. -4., the training of all training samples stops calculating after terminating；
(4) test sample data are regard as the input of the recurrence self organizing neural network after training, recurrence self organizing neural network Output be water outlet ammonia nitrogen concentration predicted value.
CN201410655729.2A 2014-11-17 2014-11-17 A kind of ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network Active CN104376380B (en)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201410655729.2A CN104376380B (en) 2014-11-17 2014-11-17 A kind of ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network
US14/668,836 US9633307B2 (en) 2014-11-17 2015-03-25 Method to predict the effluent ammonia-nitrogen concentration based on a recurrent self-organizing neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201410655729.2A CN104376380B (en) 2014-11-17 2014-11-17 A kind of ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network

Publications (2)

Publication Number Publication Date
CN104376380A CN104376380A (en) 2015-02-25
CN104376380B true CN104376380B (en) 2017-07-21

Family

ID=52555275

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201410655729.2A Active CN104376380B (en) 2014-11-17 2014-11-17 A kind of ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network

Country Status (2)

Country Link
US (1) US9633307B2 (en)
CN (1) CN104376380B (en)

Families Citing this family (35)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10539546B2 (en) * 2014-11-02 2020-01-21 Zhengbiao OUYANG Measuring phosphorus in wastewater using a self-organizing RBF neural network
CN104360035B (en) * 2014-11-02 2016-03-30 北京工业大学 A kind of sewage total phosphorus TP flexible measurement method based on self-organization population-radial base neural net
CN104965971B (en) * 2015-05-24 2017-09-01 北京工业大学 A kind of ammonia nitrogen concentration flexible measurement method based on fuzzy neural network
CN105160422B (en) * 2015-08-10 2019-03-05 北京工业大学 Sewage treatment based on self-organizing cascade neural network is discharged total phosphorus prediction technique
CN105574326A (en) * 2015-12-12 2016-05-11 北京工业大学 Self-organizing fuzzy neural network-based soft measurement method for effluent ammonia-nitrogen concentration
CN105675838B (en) * 2016-04-15 2017-12-15 北京工业大学 A based on data-driven2/ O flow water outlet total phosphorus intelligent detecting methods
CN106405030B (en) * 2016-07-27 2018-07-24 北京北排科技有限公司 A kind of water outlet total phosphorus detection device and detection method based on RBF neural
CN106295800B (en) * 2016-07-28 2019-03-05 北京工业大学 A kind of water outlet total nitrogen TN intelligent detecting method based on recurrence Self organizing RBF Neural Network
CN106872658B (en) * 2017-01-22 2019-06-18 华南理工大学 A method of the COD of sewage load prediction based on vector time series model
CN106842948B (en) * 2017-03-01 2019-08-06 长春工业大学 The method for optimally controlling of HDP molecular distillation system based on BP network
CN108793351A (en) * 2017-04-26 2018-11-13 北京济安金信科技有限公司 Sewage water treatment method, system and the electronic equipment with the system
KR102413028B1 (en) * 2017-08-16 2022-06-23 에스케이하이닉스 주식회사 Method and device for pruning convolutional neural network
CN109188903A (en) * 2018-08-20 2019-01-11 北京化工大学 The flexible measurement method of CNN furnace operation variable based on memory-enhancing effect optimization
CN109344971B (en) * 2018-09-26 2021-09-14 北京工业大学 Effluent ammonia nitrogen concentration prediction method based on adaptive recursive fuzzy neural network
CN109255200B (en) * 2018-10-23 2021-04-23 中国农业大学 Soft measurement method and device for ammonia nitrogen in aquaculture water
CN109596163B (en) * 2018-11-15 2020-12-08 北京英视睿达科技有限公司 Method and device for predicting section water quality parameter data
CN109534486B (en) * 2018-12-21 2023-07-25 哈尔滨工程大学 Ship sewage treatment control prediction system and prediction method based on random recursion wavelet neural network
CN110070220A (en) * 2019-04-16 2019-07-30 上海交通大学 A kind of ammonia nitrogen index flexible measurement method based on neural network algorithm
CN110197704B (en) * 2019-04-18 2021-06-04 北京工业大学 Self-organizing BP neural network water output total phosphorus prediction method based on limited Boltzmann machine
CN110110946B (en) * 2019-05-27 2021-09-07 江西博微新技术有限公司 Water quality prediction early warning system based on anisotropic Delaunay subdivision and implementation method thereof
CN110442991B (en) * 2019-08-12 2021-05-04 江南大学 Dynamic sulfur recovery soft measurement modeling method based on parameterized FIR (finite Impulse response) model
CN110357236B (en) * 2019-08-16 2022-01-25 江苏如是数学研究院有限公司 Sewage plant intelligent control method based on mutation inversion effluent prediction model
CN111158239B (en) * 2020-01-09 2022-06-21 上海电力大学 Association rule algorithm and neural network-based pulverizing system performance optimization method
CN111354423B (en) * 2020-02-29 2024-03-08 北京工业大学 Method for predicting ammonia nitrogen concentration of effluent of self-organizing recursive fuzzy neural network based on multivariate time series analysis
CN112114103B (en) * 2020-08-18 2021-07-20 华南理工大学 Sewage plant sludge bulking detection method based on robust adaptive canonical correlation analysis
CN112462019A (en) * 2020-11-14 2021-03-09 北京工业大学 Effluent ammonia nitrogen soft measurement method based on CL-RNN
CN114660248A (en) * 2020-12-22 2022-06-24 中国石油化工股份有限公司 COD early warning method and device based on multi-step prediction strategy
CN113029993A (en) * 2021-03-24 2021-06-25 中南大学 Ultraviolet spectrum rapid total nitrogen online detection method based on attention mechanism
CN113408205B (en) * 2021-06-22 2022-05-27 平安国际智慧城市科技股份有限公司 Environmental pollution grading method, device, equipment and storage medium
CN113723002A (en) * 2021-09-02 2021-11-30 大唐环境产业集团股份有限公司 Method and system for establishing slurry pH value prediction model of desulfurization system under all working conditions
WO2023175825A1 (en) * 2022-03-17 2023-09-21 三菱電機株式会社 Aeration amount controller and aeration amount control method
CN114626300B (en) * 2022-03-17 2023-05-02 北京工业大学 Intelligent prediction method for total nitrogen of effluent based on data discretization
CN116225083A (en) * 2022-12-13 2023-06-06 宁波市海洋与渔业研究院 Ammonia nitrogen concentration regulation and control method and system for aquaculture
CN116029589B (en) * 2022-12-14 2023-08-22 浙江问源环保科技股份有限公司 Rural domestic sewage animal and vegetable oil online monitoring method based on two-section RBF
CN117056731A (en) * 2023-09-11 2023-11-14 重庆理工大学 Ammonia nitrogen prediction method based on jet impact-negative pressure reactor flow signal

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102662040A (en) * 2012-04-20 2012-09-12 辽宁工程技术大学 Ammonian online soft measuring method for dynamic modularized nerve network
CN102854296A (en) * 2012-08-30 2013-01-02 北京工业大学 Sewage-disposal soft measurement method on basis of integrated neural network
CN103886369A (en) * 2014-03-27 2014-06-25 北京工业大学 Method for predicting effluent TP based on fuzzy neural network

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8252182B1 (en) * 2008-09-11 2012-08-28 University Of Central Florida Research Foundation, Inc. Subsurface upflow wetland system for nutrient and pathogen removal in wastewater treatment systems
US8216517B2 (en) * 2009-03-30 2012-07-10 General Electric Company System and method for monitoring an integrated system
US20150053612A1 (en) * 2012-04-27 2015-02-26 Biological Petroleum Cleaning Ltd. Method and system for treating waste material

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102662040A (en) * 2012-04-20 2012-09-12 辽宁工程技术大学 Ammonian online soft measuring method for dynamic modularized nerve network
CN102854296A (en) * 2012-08-30 2013-01-02 北京工业大学 Sewage-disposal soft measurement method on basis of integrated neural network
CN103886369A (en) * 2014-03-27 2014-06-25 北京工业大学 Method for predicting effluent TP based on fuzzy neural network

Also Published As

Publication number Publication date
US9633307B2 (en) 2017-04-25
CN104376380A (en) 2015-02-25
US20160140437A1 (en) 2016-05-19

Similar Documents

Publication Publication Date Title
CN104376380B (en) A kind of ammonia nitrogen concentration Forecasting Methodology based on recurrence self organizing neural network
CN104965971B (en) A kind of ammonia nitrogen concentration flexible measurement method based on fuzzy neural network
CN104360035B (en) A kind of sewage total phosphorus TP flexible measurement method based on self-organization population-radial base neural net
CN105510546B (en) A kind of biochemical oxygen demand (BOD) BOD intelligent detecting methods based on self-organizing Recurrent RBF Neural Networks
Han et al. Data-driven intelligent monitoring system for key variables in wastewater treatment process
CN102854296B (en) Sewage-disposal soft measurement method on basis of integrated neural network
CN104182794B (en) Method for soft measurement of effluent total phosphorus in sewage disposal process based on neural network
CN101957356B (en) Biochemical oxygen demand BOD soft measurement method based on elastic radical basis function neural network
CN103606006B (en) Sludge volume index (SVI) soft measuring method based on self-organized T-S fuzzy nerve network
CN105574326A (en) Self-organizing fuzzy neural network-based soft measurement method for effluent ammonia-nitrogen concentration
CN109492265B (en) Wastewater effluent index prediction method based on dynamic nonlinear PLS soft measurement method
CN103728431A (en) Industrial sewage COD (chemical oxygen demand) online soft measurement method based on ELM (extreme learning machine)
CN102494979B (en) Soft measurement method for SVI (sludge volume index)
CN109344971B (en) Effluent ammonia nitrogen concentration prediction method based on adaptive recursive fuzzy neural network
CN101566612A (en) Chemical oxygen demand soft-sensing method of sewage
CN107402586A (en) Dissolved Oxygen concentration Control method and system based on deep neural network
CN101923083A (en) Sewage chemical oxygen demand soft measuring method based on support vector machine and neural network
CN107664682A (en) A kind of water quality hard measurement Forecasting Methodology of ammonia nitrogen
CN110320335B (en) Multi-element robust soft measurement method for effluent quality index of sewage treatment
CN101625353A (en) Soft measurement method of outflow water quality of sewage treatment and on-line intelligent detecting instrument
CN103235096A (en) Sewage water quality detection method and apparatus
CN203772781U (en) Characteristic variable-based sewage total phosphorus measuring device
CN102778548B (en) Method for forecasting sludge volume index in sewage treatment process
CN104063609B (en) Method of assisting in judging pollution source monitoring data validity by utilizing neural network
CN107665288A (en) A kind of water quality hard measurement Forecasting Methodology of COD

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant